Chatbots for Diabetes Self-Management:

Diabetes coaching at scale

 Table of Contents
Introduction .................................................................................................................... 3
Diabetes Self-Management Landscape ..................................................................... 3
Traditional Approaches ............................................................................................. 3
Digital Approaches .................................................................................................... 4
Behavioral Shortcomings ........................................................................................... 5
Chatbots for Diabetes Self-Management ................................................................... 7
Human-like interaction and social contract ........................................................... 7
Behavior hacking and continuous learning ............................................................ 8
Chatbots are for everyone ..................................................................................... 10
Conclusion ................................................................................................................... 10

2
 Introduction
Diabetes affects 29 million people in the United states each year and costs $245
billion to treat. Successful diabetes management requires that patients create
new habits around medication adherence and glucose monitoring,
dramatically change their diets, and exercise more. Because these changes are
so difficult for people to make, fewer than 50% of patients adhere to treatment
therapies,1 contributing to the nearly 75,000 diabetes deaths per year.2

Standard approaches to improving diabetes health outcomes involve Diabetes

Self-Management Education and Support (DSME) aimed at helping patients
change behaviors and monitor their disease progression. DSME programs can
be temporarily effective; however, their effects are small and diminish over
time.3 Furthermore, because they are often highly individualized interactions
between a patient and a care provider, they are difficult to standardize, costly
to implement, and difficult to scale.

Internet-based and mobile health (mHealth) technologies have the potential to

supplement or replace standard DSME approaches. However, existing solutions
focus on only a few needs of a diabetic patient, such as nutrition guidance or
glucose monitoring,4 without considering the entirety of patient needs or how
these needs are interrelated.

Chatbots are a new and quickly developing technology that harness the
potential of mHealth while overcoming the shortcomings of existing solutions.
Chatbots rely on artificial intelligence (AI) to create conversational, interactive
experiences through a user’s SMS platform (i.e., text messaging, Facebook
messenger, etc.). Users can interact with a chatbot as they would with a friend
or family member, without requiring a downloaded mobile application. An
evidence-based behavior change chatbot could transform diabetes care by
delivering personalized yet standardized interventions that address the range of
patients’ needs at scale.

Diabetes Self-Management Landscape

Traditional Approaches

The American Diabetes Association includes DSME as a core component of

diabetes care. DSME provides individualized education and support to patients
with the goal of preventing complications, slowing disease progression, and
improving health outcomes. DSME addresses both lifestyle and psychosocial
issues and helps patients manage monitoring and medication regimens.

3
Depending on specific patient needs, DSME can help patients lose weight,
change their diet, increase physical activity, quit smoking, and build habits
around monitoring blood glucose, taking medication, and checking for signs of
complications.

DSME is typically delivered through one-on-one or group-based counseling with

a nurse, dietician, or certified diabetes coach. Programs last for an average of 6
months with around 18 hours of contact between the patient and the care
provider.5 Programs may also include phone support, text communications, or
educational tools.

Programs are very focused on goals that are set collaboratively between the
patient and the care team. The National Standards stipulate that programs are
individualized, so patients are encouraged to come up with their own goals.
National Standards also include monitoring patient progress, so clinical
outcomes are tracked closely.6

There may be some benefits from lifestyle interventions delivered through DSME
programs; however, in practice, many lifestyle interventions place undue
emphasis on health outcomes (like weight loss) and have limited clinical benefit.
Positive impacts are inconsistent across the different types of interventions,7 and
when effects are found, they are small and become smaller over time.8
Furthermore, interventions typically do not impact diabetes risk outcomes or
mortality at all.9

Diabetes is often associated with comorbidities or difficult lifestyle

circumstances, such as financial hardship, food insecurity, or lack of social
support, increasing the difficulty of self-management. The specific needs of
individual diabetic patients are wide-ranging and highly variable. Traditional
diabetes management programs have addressed these diverse needs by
providing highly individualized care programs that require intensive one-on-one
interactions between patients and providers or well-coordinated care provider
teams. These interactions allow for personalized care, but at a cost: researchers
and health providers are unable to identify the specific elements of interventions
that really work.10 As a result, DSME programs are difficult to standardize and
scale, costly, and inaccessible to patients who need support, which may
explain why only 5-7% of eligible patients participate in them.11,12

Digital Approaches

Mobile health (mHealth) and internet tools have become increasingly popular
alternatives, or additions, to traditional DSME programs. mHealth in particular
has great potential to expand care delivery as smartphones are ubiquitous and

4
can integrate data from a patient’s numerous connected devices (e.g.
glucometer or Fitbit).

Today’s apps claim to assist in managing different components of diabetes care

with features such as the ability to track glucose, insulin, medication, diet,
and/or exercise.13 For example, MySugr, which has nearly 900,000 downloads,
combines data from multiple devices and allows for meal tracking so users can
view trends and share data with their physicians. Other less common features in
diabetes apps include reminders for refills, data logs that can be uploaded to
care providers, medication information,14 and calendar-based alarms to remind
patients to take medications.15

The majority of existing apps focus on only a few needs of a diabetic patient
(e.g., nutrition guidance or glucose monitoring),16 and few focus on addressing
the behavioral or psychological barriers to successful disease management.
Consequently, there is little evidence that these apps improve health
outcomes17 or are a cost-effective approach to delivering DSME.18

Behavioral Shortcomings

Both traditional and mHealth solutions fail to take into account behavioral
perspectives that could greatly improve patient self-management. National
Standards include behavioral goals as an important component for DSME
programs,19 but ‘behavioral goals’ are not clearly defined. Furthermore, the
examples cited to support this component of DSME do not take into account
behavioral science evidence around creating new routines and habits or
overcoming psychological barriers.

mHealth solutions also lack a behavioral science foundation to their design and
instead are overly focused on tracking and trend reporting as mechanisms for
behavior change.20 For example, the MySugr app is centered on tracking and
viewing data trends. Users can track glucose, insulin, carbs, and exercise, and
will estimate HbA1c values if users log their blood glucose values three times a
day for seven days.

5
 MySugr Dashboard

Tracking features are typically evaluated favorably and patients believe that
viewing this data improves health.21 However, these ratings and beliefs about
efficacy should not be confused for true drivers of behavior change. While some
forms of self-monitoring, such as recording diet and exercise, can be effective in
certain cases,22 emerging evidence suggests that showing a patient trends in
their clinical values is not effective at changing behavior.23

This overreliance on presenting data back to patients may also have

unintended consequences. Patients with low numeracy or cognitive bandwidth
may become overburdened by this data feedback and disengage from the
app altogether. Patients may misinterpret normal fluctuations in their data as
meaningful trends and erroneously associate these changes with whatever
behaviors are salient in the moment. This tendency would make it more difficult
to learn how and why specific behaviors impact health.

A more effective solution is one that incorporates evidence from behavioral

science around how to change and sustain behaviors along with the

6
individualized care of traditional DSME programs, while taking advantage of the
scalability of mHealth.

Chatbots for Diabetes Self-Management

Chatbot technology has enormous potential for transforming DSME and
changing patient behaviors. Artificial intelligence (AI), natural language
processing (NLP), and machine learning (ML) allow chatbots to have
conversational interactions with the user, interpret meaning from the user’s
responses (such as emotional valence or time preferences), and learn from prior
interactions with the user.

From the user’s perspective, the bot is seen as an expert in a particular domain,
with a personality and human-like characteristics. Over time, as the user
interacts with the bot in a social manner, the user becomes attached to the bot
in much the same way they would with a friend or family member. This social
relationship creates accountability and commitment between the bot and the
user. The bot is responsible for the user’s needs and the user is therefore
compelled to meet the bot’s expectations.

Interacting with a chatbot can feel like interacting a human, except that a
chatbot is much smarter than a human. Behind the scenes, a chatbot app can
integrate data from any connected device, triggering an interaction in
response to flags in the user’s data. The chatbot can provide continuous and
reliable support to the user in an experience that is customized to each patient
but standardized based on predetermined parameters.

As evidence of their unique value, chatbot apps are rapidly emerging to

address health problems. For example, Emile offers to help users sleep better
and exercise. Emile promises to “never give up on you.” X2AI provides mental
health care, like cognitive behavioral therapy, in places where people would
not otherwise have access. These products are in their early phases of
development, but their use cases are expanding rapidly.

Human-like interaction and social contract

Chatbots are unlike other technology solutions because they are personified.
They feel like people so we treat them like people. A chatbot has a name, the
interactions are conversational and responsive, and the experience exists within
the same platforms where users interact with friends and family.

Even when we know we’re interacting with a computer, we often

anthropomorphize, or assign human characteristics to robots.24 We apply the

7
same mental models to these human-like interactions that we apply to humans:
we create social contracts, we feel trust and empathy, and we feel social
pressure to not disappoint. Therefore, we can rely on a robot, or chatbot, the
same way we rely on people. We assume they have expectations of us, just as a
friend, family member, or therapist might.

With traditional DSME programs, the more interactions between a patient and a
care provider, the more effective the intervention.25 This is likely because
increased interactions leads to stronger social relationships, which provides
social accountability for the patient’s behaviors. Social interactions are also
salient and meaningful to us, so they may have a greater influence on our
behavior compared to non-social experiences. Chatbots show great promise for
delivering DSME at scale because they can recreate the human element of
these interactions that is likely so powerful for patients.

Social accountability, provided by a chatbot

Behavior hacking and continuous learning

Because a chatbot engages in conversational interactions with a user, there is

an opportunity to collect rich data from users. For example, a chatbot could ask

8
all users a set of behavioral diagnostic questions periodically over time, which
would then inform the type of behavioral interventions the chatbot would
deliver to the user. Users who report having fewer daily habits could be exposed
to interactions with the chatbot that are focused on building new habits,
whereas users with more negative affective responses could be exposed to
interventions focused on mindfulness or other emotion regulation strategies.

In a downloaded app, a user’s experience is constrained to a set of

predetermined interactions or the content that is included in the app. With a
chatbot, the user can also initiate interactions across a wider range of topics.
For example, a user could ask the chatbot about specific symptoms as they are
occurring. The bot would use natural language processing to interpret the user’s
unprompted question and provide an appropriate response.

Conversational alert and symptom feedback, from a chatbot

A chatbot app has the potential to learn from the user and change the user’s
experience based on previous interactions and the user’s behavior. A chatbot
app could also learn from all of the users and recognize patterns, identify types
of users who respond particularly well to certain types of prompts or

9
interventions, and eventually automatically and precisely funnel users to the
most effective journey or behavioral intervention for them.

Chatbots are for everyone

Chatbots on mobile devices work within existing messaging platforms like

iMessage or WhatsApp, and do not require the user to download anything new.
Users simply install the chatbot by providing their phone number or sending a
text message. Users do not need to learn how to use a new app or build new
routines around the new app, since the interaction with the chatbot exists within
platforms where they are already regularly engaged, such as text messaging.

Messaging-based apps also have substantially slower drop-off compared to

standard apps, leading to continued use and greater long-term retention. The
interactions do not need to be predefined, resulting in more engaging and
varied experiences, and a higher likelihood of health improvements.26 Chatbot
apps do not require users to have smartphones, either, which means they are
more accessible to low-income and elderly populations who are less likely to
have smartphones but have other mobile devices.

Conclusion
Diabetes impacts a large number of Americans each year. Medical and
behavioral treatments are effective but only when patients successfully and
permanently change their behaviors.

Mobile technology offers enormous opportunity for integrating into patients’

everyday lives in ways that enable long-term behavior change. However, the
existing smartphone apps have limited functionality and are not suitable for a
large segment of the population. Chatbots, implementing artificial intelligence
through SMS platforms, offer enormous potential to take advantage of the
benefits of both mobile technology and the power of human interactions to
offer a scalable, cost-effective, and powerful diabetes behavior change
intervention.

1World Health Organization: Adherence to long-term therapies. Evidence for action. Geneva: World Health
Organization; 2003
2Kochanek KD, Murphy SL, Xu J, et al. Deaths: Final data for 2014. National vital statistics reports; vol 65 no 4.

Hyattsville, MD: National Center for Health Statistics. 2016.

3 Norris, S. L., Lau, J., Smith, S. J., Schmid, C. H., & Engelgau, M. M. (2002). Self-Management education for

adults with type 2 Diabetes A meta-analysis of the effect on glycemic control. Diabetes care, 25(7), 1159-
1171.

10

4 Hood, M., Wilson, R., Corsica, J., Bradley, L., Chirinos, D., & Vivo, A. (2016). What do we know about mobile
applications for diabetes self-management? A review of reviews. Journal of behavioral medicine, 39(6),
981-994.
5 Chrvala, C. A., et al. (2016). Diabetes self-management education for adults with type 2 diabetes mellitus:

A systematic review of the effect on glycemic control. in Patient Education and Counseling journal.
6 Haas, L., Maryniuk, M., Beck, J., Cox, C. E., Duker, P., Edwards, L., ... & McLaughlin, S. (2012). National

standards for diabetes self-management education and support. The Diabetes Educator, 38(5), 619-629.
7 Chen, L., Pei, J. H., Kuang, J., Chen, H. M., Chen, Z., Li, Z. W., & Yang, H. Z. (2015). Effect of lifestyle

intervention in patients with type 2 diabetes: a meta-analysis. Metabolism, 64(2), 338-347.

8 Norris, S. L., Lau, J., Smith, S. J., Schmid, C. H., & Engelgau, M. M. (2002). Self-management education for

adults with type 2 diabetes. Diabetes care, 25(7), 1159-1171.

9 Schellenberg, E. S., Dryden, D. M., Vandermeer, B., Ha, C., & Korownyk, C. (2013). Lifestyle interventions for

patients with and at risk for type 2 diabetes: a systematic review and meta-analysis. Annals of internal
medicine, 159(8), 543-551.
10 Schinckus, L., Van den Broucke, S., Housiaux, M., & Diabetes Literacy Consortium. (2014). Assessment of

implementation fidelity in diabetes self-management education programs: a systematic review. Patient

education and counseling, 96(1), 13-21.
11 Strawbridge LM, Lloyd JT, Meadow A, Riley GF, Howell BL. Use of Medicare’s diabetes self- management

training benefit. Health Educ Behav 2015;42:530–538

12 Li R, Shrestha SS, Lipman R, Burrows NR, Kolb LE, Rutledge S; Centers for Disease Control and Prevention

(CDC). Diabetes self-management education and training among privately insured persons with newly
diagnosed diabetes–United States, 2011-2012. MMWR Morb Mortal Wkly Rep 2014;63:1045–1049
13 Hood, M., Wilson, R., Corsica, J., Bradley, L., Chirinos, D., & Vivo, A. (2016). What do we know about

mobile applications for diabetes self-management? A review of reviews. Journal of behavioral medicine,
39(6), 981-994.
14 Breland, Jessica Y., Vivian M. Yeh, and Jessica Yu. "Adherence to evidence-based guidelines among

diabetes self-management apps." Translational behavioral medicine 3.3 (2013): 277-286.

Tran, Joseph, Rosanna Tran, and John R. White. "Smartphone-based glucose monitors and applications in
the management of diabetes: an overview of 10 salient “apps” and a novel smartphone-connected blood
glucose monitor." Clinical Diabetes 30.4 (2012): 173-178.
Georga, Eleni I., et al. "Wearable systems and mobile applications for diabetes disease management."
Health and Technology 4.2 (2014): 101-112.
15 Yuan, Shupei, et al. "Keep using my health apps: Discover users' perception of health and fitness apps

with the UTAUT2 model." Telemedicine and e-Health 21.9 (2015): 735-741.
Lyons, Elizabeth J., et al. "Behavior change techniques implemented in electronic lifestyle activity monitors:
a systematic content analysis." Journal of medical Internet research 16.8 (2014): e192.
Eng, Donna S., and Joyce M. Lee. "The promise and peril of mobile health applications for diabetes and
endocrinology." Pediatric diabetes 14.4 (2013): 231-238.
Brzan, P. P., Rotman, E., Pajnkihar, M., & Klanjsek, P. (2016). Mobile applications for control and self
management of diabetes: A systematic review. Journal of medical systems, 40(9), 210.
16 Hood, M., Wilson, R., Corsica, J., Bradley, L., Chirinos, D., & Vivo, A. (2016). What do we know about

mobile applications for diabetes self-management? A review of reviews. Journal of behavioral medicine,
39(6), 981-994.
17 Holtz, B., & Lauckner, C. (2012). Diabetes management via mobile phones: a systematic review.

Telemedicine and e-Health, 18(3), 175-184.

18 Shah, Viral N., and Satish K. Garg. "Managing diabetes in the digital age." Clinical Diabetes and

Endocrinology 1.1 (2015): 1.

19 Haas, L., Maryniuk, M., Beck, J., Cox, C. E., Duker, P., Edwards, L., ... & McLaughlin, S. (2012). National

standards for diabetes self-management education and support. The Diabetes Educator, 38(5), 619-629.
20 Brzan, P. P., Rotman, E., Pajnkihar, M., & Klanjsek, P. (2016). Mobile applications for control and self

management of diabetes: A systematic review. Journal of medical systems, 40(9), 210.

21 PWC ‘The Wearable Life 2.0; Connected living in a wearable world”,

http://www.pwc.com/us/en/industry/entertainment-media/assets/pwc-cis-wearables.pdf
22 Burke, L. E., Wang, J., & Sevick, M. A. (2011). Self-monitoring in weight loss: a systematic review of the

literature. Journal of the American Dietetic Association, 111(1), 92-102.

23 Jakicic, J. M., Davis, K. K., Rogers, R. J., King, W. C., Marcus, M. D., Helsel, D., ... & Belle, S. H. (2016). Effect

of wearable technology combined with a lifestyle intervention on long-term weight loss: the IDEA
randomized clinical trial. Jama, 316(11), 1161-1171.

11

24 Broadbent, E., Kumar, V., Li, X., Sollers 3rd, J., Stafford, R. Q., MacDonald, B. A., & Wegner, D. M. (2013).
Robots with display screens: a robot with a more humanlike face display is perceived to have more mind
and a better personality. PloS one, 8(8), e72589.
25 Tang, T. S., Funnell, M. M., Brown, M. B., & Kurlander, J. E. (2010). Self-management support in “real-world”

settings: an empowerment-based intervention. Patient education and counseling, 79(2), 178-184.

26 Cole-Lewis, H., & Kershaw, T. (2010). Text messaging as a tool for behavior change in disease prevention

and management. Epidemiologic reviews, 32(1), 56-69.

12